Академический Документы
Профессиональный Документы
Культура Документы
Acknowledgement
The authors would like to thank for reviewing the guide: Thomas Reichel (Technical consultant), Blair Hall
(MSL), Gary Bennett (National Instruments), Dave Blackham (Keysight Technologies), Andreas C. Böck
(esz AG calibration & metrology), Andy Brush (TEGAM), Tekamul Buber (Maury Microwave), Albert Calvo
(Rohde & Schwarz), Onur Cetiner (Keysight Technologies), Chris Eio (NPL), Andrea Ferrero (Keysight
Technologies), Israel Garcia Ruiz (CENAM), Martin Grassl (Spinner), Tuomas Haitto (Millog Oy), Johannes
Hoffmann (METAS), Matthias Hübler (Rohde & Schwarz), Ian Instone (Technical consultant), Harald Jäger
(Rohde & Schwarz), Karsten Kuhlmann (PTB), Jian Liu (Keysight Technologies), Linoh Magalula (NMISA),
Jon Martens (Anritsu), Guillermo Monasterios (INTI), Faisal Mubarak (VSL), Rusty Myers (Keysight
Technologies), Reiner Oppelt (Rosenberger), Nick Ridler (NPL), Juerg Ruefenacht (METAS), Handan
Sakarya (UME), Bart Schrijver (Keysight Technologies), Joachim Schubert (Rosenberger), Nosherwan
Shoaib (INRIM, NUST), Hernando Silva (INTI), Pamela Silwana (NMISA), Laszlo Sleisz (NMHH), Daniel
Stalder (METAS), Michael Wollensack (METAS), Ken Wong (Keysight Technologies), Sherko Zinal (PTB).
EURAMET e.V.
Bundesallee 100
38116 Braunschweig
Germany
E-mail: secretariat@euramet.org
Phone: +49 531 592 1960
Official language
The English language version of this document is the definitive version. The EURAMET Secretariat can
give permission to translate this text into other languages, subject to certain conditions available on
application. In case of any inconsistency between the terms of the translation and the terms of this
document, this document shall prevail.
Copyright
The copyright of this document (EURAMET Calibration Guide No. 12, version 3.0 – English version) is held
by © EURAMET e.V. 2007. The text may not be copied for sale and may not be reproduced other than in
full. Extracts may be taken only with the permission of the EURAMET Secretariat.
ISBN 978-3-942992-51-0
Guidance publications
This document gives guidance on measurement practices in the specified fields of measurements. By
applying the recommendations presented in this document, laboratories can produce calibration results
that can be recognised and accepted throughout Europe. The approaches taken are not mandatory and
are for the guidance of calibration laboratories. The document has been produced as a means of promoting
a consistent approach to good measurement practice leading to and supporting laboratory accreditation.
The guide may be used by third parties, e.g. National Accreditation Bodies, peer reviewers, witnesses to
measurements, etc., as a reference only. Should the guide be adopted as part of a requirement of any such
party, this shall be for that application only and the EURAMET Secretariat should be informed of any such
adoption.
No representation is made nor warranty given that this document or the information contained in it will be
suitable for any particular purpose. In no event shall EURAMET, the authors or anyone else involved in the
creation of the document be liable for any damages whatsoever arising out of the use of the information
contained herein. The parties using the guide shall indemnify EURAMET accordingly.
Further information
For further information about this document, please contact your national contact person in the EURAMET
Technical Committee for Electricity and Magnetism (see https://www.euramet.org)
Purpose
This document has been produced to enhance the equivalence and mutual recognition of
calibration results obtained by laboratories performing measurements with vector network
analysers.
List of Figures 6
List of Tables 7
1 Introduction 9
1.1 Purpose of this guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Comparison with previous guideline . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Scope and Applicability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 VNA calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.2 Error model and measurement model . . . . . . . . . . . . . . . . . . . 11
1.4.3 Error coefficients and residual errors . . . . . . . . . . . . . . . . . . . . 11
3 VNA calibration 13
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Practical advice for VNA calibration . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Electronic calibration units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 Verification 15
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Purpose of verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3 Verification Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.1 Coincidence tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.2 Plausibility tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 Verification procedure and practical advice . . . . . . . . . . . . . . . . . . . . . 16
4.5 Verification criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Uncertainty contributions 18
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Identification of influence quantities . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Characterization of uncertainty contributions . . . . . . . . . . . . . . . . . . . . 19
5.3.1 Characterization of calibration standards . . . . . . . . . . . . . . . . . . 19
5.3.2 Noise floor and trace noise . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3.3 VNA non-linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3.4 VNA drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3.5 Isolation (cross-talk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.6 Test port cable stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3.7 Connection repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6 Measurement model 23
References 33
Annexes 39
A Glossary 39
B Notation 43
D VNA calibration 51
D.1 One-port calibration techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
D.2 Two-port calibration techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.2 Ten-term error model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
D.2.3 Seven-term error model . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
D.3 Over-determined calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
H Ripple Method 91
H.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.2 Uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.3 Practical Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
H.4 Measurement model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
H.4.1 One-port equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
H.4.2 Two-port equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
H.5 Uncertainty contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
H.5.1 Directivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
H.5.2 Source match . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
H.5.3 Reflection tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
H.5.4 Transmission tracking and load match . . . . . . . . . . . . . . . . . . . 100
H.5.5 Isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
H.5.6 Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
H.5.7 Test port cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
H.5.8 Non-linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
H.5.9 Repeatability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
J Examples 107
J.1 One-port matched load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
J.2 One-port mismatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
J.3 One-port short . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
J.4 Two-port Adapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
J.5 Two-port 20 dB attenuation device . . . . . . . . . . . . . . . . . . . . . . . . . 122
J.6 Two-port 50 dB attenuation device . . . . . . . . . . . . . . . . . . . . . . . . . 128
List of Tables
8.1 Recommended minimal pin gaps of coaxial connections . . . . . . . . . . . 32
D.1 Error coefficients of one-port error model . . . . . . . . . . . . . . . . . . . . 51
D.2 Error coefficients of ten-term error models . . . . . . . . . . . . . . . . . . . 53
D.3 Error coefficients of seven-term error models . . . . . . . . . . . . . . . . . 55
F.1 Influences in one-port measurement model . . . . . . . . . . . . . . . . . . 68
J.1 Uncertainty budget with rigorous method for Re (S11 ) of matched load . . . 111
J.2 Uncertainty budget with rigorous method for Im (S11 ) of matched load . . . 111
J.3 Uncertainty budget with Ripple Method for S11 of matched load . . . . . . . 111
J.4 Uncertainty budget with rigorous method for |S11 | of mismatch . . . . . . . . 113
J.5 Uncertainty budget with rigorous method for arg (S11 ) of mismatch . . . . . 113
J.6 Uncertainty budget with Ripple Method for S11 of mismatch . . . . . . . . . 113
J.7 Uncertainty budget with rigorous method for |S11 | of a short . . . . . . . . . 115
J.8 Uncertainty budget with rigorous method for arg (S11 ) of a short . . . . . . . 115
J.9 Uncertainty budget with Ripple Method for S11 of a short . . . . . . . . . . . 115
J.10 Uncertainty budget with rigorous method for Re (S22 ) of an adapter . . . . . 117
J.11 Uncertainty budget with rigorous method for Im (S22 ) of an adapter . . . . . 117
J.12 Uncertainty budget with Ripple method for S22 of an adapter . . . . . . . . . 118
J.13 Uncertainty budget with rigorous method for |S12 | of an adapter . . . . . . . 120
J.14 Uncertainty budget with rigorous method for arg (S12 ) of an adapter . . . . . 120
J.15 Uncertainty budget with Ripple method for S12 of an adapter . . . . . . . . . 121
J.16 Uncertainty budget with rigorous method for Re (S22 ) of a 20 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
J.17 Uncertainty budget with rigorous method for Im (S22 ) of an adapter . . . . . 123
J.18 Uncertainty budget with Ripple method for S22 of a 20 dB attenuation device 124
J.19 Uncertainty budget with rigorous method for |S12 | of a 20 dB attenuation device 126
J.20 Uncertainty budget with rigorous method for arg (S12 ) of a 20 dB attenuation
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
J.21 Uncertainty budget with Ripple method for S12 of a 20 dB attenuation device 127
1.1.2 While this guide is primarily intended for accredited calibration laboratories it is as well
useful for national metrology institutes and anybody else performing VNA measure-
ments.
1.1.3 This guide explains how to establish SI (international system of units) traceability using
reference standards (section 2). It gives advice on VNA calibration schemes (section 3)
and verification (section 4). It has a particular focus on the evaluation of measurement
uncertainties (sections 5, 6 and 7). Finally, advice on good measurement practice is
given (section 8).
1.1.4 It was the intent to keep the main body of this guide reasonably short with a focus on
practical aspects to make it useful for the practitioner. The appendices contain further
details on many of the topics covered in the main body of the document. Examples
are given to illustrate some of the concepts. A glossary is provided and references are
cited for further reading.
1.2.2 The previous versions of the guide promoted SI traceability through beadless air-
dielectric lines using the so-called Ripple Method. This method only partly relies on
a measurement model and makes some questionable assumptions. It has a potential
to either underestimate or overestimate the measurement uncertainty and is limited in
applicability. At higher frequencies it becomes unpractical.
1.2.3 In 2011, Supplement 2 [5] of the ISO-GUM [6] has been published giving thorough
advice on how to evaluate measurement uncertainties for multi-dimensional measur-
ands, including complex-valued quantities. It is therefore the authoritative guideline for
the determination of measurement uncertainties associated with S-parameters. In par-
allel, software capabilities have become available to support the sometimes elaborate
calculation of the measurement uncertainty.
1.2.5 This guide gives advice on how to implement the rigorous propagation of uncertainty
to achieve best accuracy in VNA measurements. Due to the widespread use of the
Ripple Method, it is still included in the guide in an improved form, by clearly specifying
its limits in terms of stated uncertainty and applicability.
1.2.6 Compared to previous versions, this guide provides additional advice on VNA mea-
surement models, VNA calibration, VNA verification, and best measurement practice.
Furthermore, it provides an introduction to measurement uncertainties associated with
complex-valued quantities.
1.4 Terminology
Some of the terms used in VNA metrology are rooted in history but are not coherent
with contemporary terminology used in metrology. The subsequent clauses briefly
address these issues.
2.1.2 The first fundamental step of the traceability chain of S-parameters is established with
the characterization of calculable measurement standards [9, 10, 11, 12]. Known cal-
culable coaxial standards are air-dielectric lines, offset shorts, flush shorts and offset
opens. Calculable standards are parametrized and measured dimensionally. The me-
chanical model should cover the whole standard including the connector interface to
achieve highest accuracy and to obtain a consistent definition of the measurement
reference plane [13, 14]. Based on dimensional measurements and known material
parameters the S-parameters of the calculable standards are determined with the help
of analytical equations and numerical EM simulations in combination with electrical
measurements. Calculable standards may be called primary standards. The process
of establishing traceability by using primary standards is called a primary experiment.
Primary experiments are elaborate and require measurement and modeling capabili-
ties at a level that can usually only be provided by national metrology institutes.
2.1.3 A VNA being calibrated by using primary standards can be used to characterize other
calibration standards to disseminate S-parameter traceability to lower levels in the
2.2.6 Some of the calibration and verification standards might be of the same type but they
must be physically different items. A single standard must not be used for calibration
and verification simultaneously. The only exception is the re-measurement of the cal-
ibration standards after calibration to test the stability of the measurement system, as
described in 4.3.2.
2.2.7 The memory of older VNAs is limited in size. To cope with this limitation, manufac-
turers of calibration standards used a polynomial representation [3, 15] to model the
frequency response of the highly reflective calibration standards. In addition it is as-
sumed that the matched load is ideal. The measurement accuracy is directly limited
by the quality of the polynomial curve fit, representing the highly reflective standards,
and how well the matched load approximates the reference impedance, e.g. 50 Ω. At
higher frequencies, typically above 2 GHz, a sliding load is used instead of a fixed load
since it provides superior matching performance.
2.2.8 Modern VNAs are able to make use of data-based standard definitions. The S-parameters
of the standards and associated uncertainties are stored as data sets with reasonably
high frequency resolution. Uncertainties associated with the characterization of the
standards determine the measurement accuracy of the calibrated VNA. Measurement
accuracy is not dependent on the fixed load being matched as perfectly as possible.
Thus the use of a sliding load is dispensable. It is still necessary that the standards
behave in a “normal” way, i.e. that the fixed load offers a reasonable matching and the
highly reflective standards principally act as reflecting devices. For best measurement
accuracy, it is essential that the S-parameters of the standards, at any frequency, do
not cluster in a single location of the complex measurement plane.
3 VNA calibration
3.1 Introduction
3.1.1 VNA calibration is the process of determining the VNA systematic measurement errors,
which are called error coefficients (sometimes also referred to as error terms or cali-
bration coefficients). The process is based on a VNA error model. A VNA error model
relates the values indicated by the VNA (raw S-parameters) to the S-parameters of the
device connected to the test port of the VNA. Part of this mathematical relationship are
the unknown error coefficients. During VNA calibration, a set of known calibration stan-
dards is measured and the error coefficients are determined. They can then be used to
transform the raw S-parameters of a DUT measurement into corrected S-parameters,
a step which is usually being referred to as VNA error correction.
3.1.2 VNA error models are based on the idea that the signal flow in a VNA can be modeled
as a linear network. The VNA error model can be illustrated graphically by a signal flow
graph. There are methods for reducing signal flow graphs into mathematical equations,
as e.g. described in [16, 17]. These mathematical equations are used to perform VNA
calibration and VNA error correction.
3.1.4 Since the development of computer-controlled VNAs, several calibration methods have
been developed. The choice between them depends on the final application, the num-
ber and the type of the calibration standards and the desired accuracy. Some details
on one-port and two-port calibration schemes can be found in appendix D.
3.2.2 A useful alternative to SOLT is the SOLR (short - open - load - reciprocal, also known
as Unknown Thru) calibration [18]. It uses measurements of short, open and matched
load at each port and the reciprocity condition in the transmission measurement of the
DUT to determine the error coefficients. SOLR should be applied in situations where
the measurement of a flush thru is not practical. This includes cases where the DUT is
of extended size, i.e. a relatively large cable movement would be necessary between
the measurement of the flush thru during calibration and the subsequent measurement
of the DUT. It further includes measurements of non-insertable DUTs and DUTs that
have different connector types and/or different impedance definitions at each port. It
should be noted that applying the SOLR calibration usually requires a four-receiver
architecture of the VNA. Compared to SOLT, SOLR calibration is more sensitive to the
characterization of short and open calibration standards. For further details on VNA
architecture, SOL, SOLT and SOLR refer to appendix D.
3.2.3 Using more calibration standards than needed results in an over-determined calibration
algorithm. For details see D.3. This type of calibration is useful to enhance the accuracy
of the calibration and to detect problems with single calibration standards. The software
(VNA firmware or external program), however, has to to support this type of calibration.
4 Verification
4.1 Introduction
According to the VIM [7], verification generally denotes provision of objective evidence
that a given item fulfills specified requirements. In the present case of VNA measure-
ments, the purpose of verification is to confirm that the VNA, test port cables, and
calibration standards are working together correctly as a measurement system, i.e.,
that the stated measurement uncertainty is met. This section has a focus on practical
advice related to verification. More details are provided in appendix E.
4.3.2.2 In the most common case of the reflection coefficient S11 of mechanical offset opens
and offset shorts, the frequency response loci should produce approximately circular
arcs on the Smith Chart edge approaching the reflection coefficient S11 = 1 point
(right-hand side) for the opens and S11 = −1 (left-hand side) for the shorts for low
(MHz) frequencies, respectively.
4.3.2.3 The measurement of a flush short is suitable to test how well the reference plane has
been defined by VNA calibration. As a flush short is a passive device it is expected
that the magnitude of the reflection coefficient will not exceed |S11 | = 1. A perfect
flush short would show no phase change with frequency. Due to imperfections in the
fabrication process this is, however, never the case. In reality the phase should show
a slow monotonous clockwise change starting at -180 deg. Any deviation from this
behavior is an indication that the reference plane has not been properly defined during
VNA calibration, see e.g. [13].
4.3.2.4 Applying different VNA calibration algorithms, preferably with different sets of calibra-
tion standards, should generally lead to equivalent results for the measurements of
the DUTs, conditional to the associated measurement uncertainties. An effective and
also quantitative test is the comparison of SOLT and SOLR calibration, as discussed
in E.1.2.6.
4.3.2.5 After an SOLT calibration it is useful to measure a reciprocal two-port device with finite
electrical length to verify the symmetry, i.e. S21 = S12 . Any significant deviation would
indicate problems related to the calibration.
4.3.2.6 If the frequency range is divided up into bands, and different sets of calibration stan-
dards or different calibration methods are used, then measurement results should show
fairly continuous transitions when crossing between frequency bands.
4.4.3 Verification and calibration standards that have been calibrated by the same laboratory
can potentially be used to apply sharper verification criteria, see 4.5.6. It is therefore
recommended but not necessary.
4.4.4 As a general rule it is advised to perform a coincidence test using at least one verifica-
tion standard having a response similar to the DUT.
4.4.5 For a more general verification covering the entire region of the Smith Chart the verifi-
cation should be performed as a two-step procedure. First, the one-port performance
of all ports has to be checked by measuring calibrated one-port verification standards.
This is followed by the two-port verification.
4.4.6 For the one-port verification it is recommended to measure two highly reflective stan-
dards with different phase (different offset lengths compared to calibration standards)
and one matched load. The matched load might be replaced by a matched attenuation
device with ≥30 dB of attenuation. To check the non-linearity of the measurements it
is also recommended to measure a mismatch standard with a reflection coefficient of
approximately 0.2 to 0.5.
4.4.7 For the two-port verification a transparent device (beaded air-dielectric line or adapter)
and a set of matched attenuation devices should be measured. Most manufacturer
verification kits contain a 20 dB and a 40 dB attenuation device. It is recommended to
perform an additional measurement using a 3 dB attenuation device to have a better
coverage of the Smith Chart and to detect problems related to non-linearity.
4.5.3 Visual inspection of the frequency response of the data, i.e. the difference between
measurement and characterization data with associated uncertainties, is highly rec-
ommended. It allows a judgement to be made about the quality of the verification
4.5.4 Quantitative evaluation of verification data requires mathematical analysis of the data
and the associated uncertainties. Automation of data analysis might be useful if a large
amount of data has to be evaluated. A method for quantitative evaluation of verification
data is shown in E.2.
4.5.6 If calibration standards and verification standards have been calibrated by the same
laboratory, there is likely to be a strong positive correlation between standards of the
same type, e.g. the matched load used for calibration and the matched load used for
verification. Consequently this will lead to a similar correlation between measurement
and reference data of the verification standard. This will further lead to a smaller un-
certainty associated with the difference between measurement and reference data and
therefore to a sharper pass/fail decision. However, this can only be exploited, if the
correlation information between calibration and verification standards is available and
if the evaluation of the uncertainty associated with the measurement of the verification
standards considers this correlation information to the full extent. Obviously, quanti-
tative evaluation according to the formalism in E.2 is needed to take correlation into
account.
5 Uncertainty contributions
5.1 Introduction
This section identifies and discusses quantities influencing VNA measurements. There
are a number of quantities with significant influence on the result of a VNA measure-
ment. The measurement models discussed in section 6 refer to these quantities ex-
plicitly. Estimates of these quantities, and the uncertainties associated, are inputs to
the measurement models to determine a value and uncertainty associated with the
quantity of interest, usually the S-parameters of the DUT.
• VNA drift
• Isolation (cross-talk)
• Connection repeatability
In some specific cases other VNA parameters such as frequency or output power have
to be considered in the uncertainty analysis too. However, this is rather unusual and
therefore not considered here (see also 2.1.5).
5.2.2 All these influences are sources of measurement errors. It is therefore essential that
they be part of any model that describes the measurement and that is being used to
evaluate the uncertainty of the measurement. Any uncertainty evaluation that does not
address all the components listed above has to be considered as incomplete.
5.3.1.1 Characterized calibration standards are necessary if the uncertainty evaluation is per-
formed with the rigorous method, according to 7.1. If the Ripple Method is applied,
according to 7.2, it is also possible just to use manufacturer data, which is often pro-
vided along with the calibration kit.
5.3.1.2 The reference data of calibration standards with associated uncertainties are taken
from a calibration certificate, unless the laboratory has its own realization of traceability
to SI units. Normally, this is only the case for national metrology institutes.
5.3.1.3 Usually the dominant source of uncertainty in a VNA measurement are associated with
characterisations of calibration standards. Accurate characterization of the calibration
standards includes both body and connector [13]. For an accurate determination of
the measurement uncertainty it is preferable that correlations between the uncertainty
components of the calibration standards are taken into account. Neglecting correla-
tions can lead to either underestimation or overestimation of the measurement uncer-
tainty.
5.3.1.5 Calibration standards can drift or change unexpectedly over time. A slow drift due to
aging is normal and it is therefore necessary to recalibrate the standards on a regular
basis. The recalibration interval depends on the operating conditions and needs to be
defined on a case to case basis. Unusual drift or change can be detected through
verification, see section 4. Good care and best measurement practice, see section 8,
are advised to minimize aging. Aprupt changes or drift in the calibration standards can
also be detected by using more calibration standards than necessary and performing
an over-determined calibration, see D.3.
5.3.2.1 Noise denotes random signal fluctuations, which are characteristic of all electronic
circuits. For VNA measurements two types of noise are distinguished. Noise floor
denotes random fluctuations in the absence of a deterministic signal. Trace noise de-
notes random fluctuations of the measurement result. When trace noise is far above
the noise floor, it is proportional to the measured value, i.e., it is a constant fraction
of the result. In absolute notation, however, trace noise grows in proportion to an in-
creasing measurement result. The noise content of a transmission coefficient can thus
be characterized by a single dB value, as far as the contribution from the noise floor
is negligible. Noise floor and trace noise are dependent on the setting of the VNA, for
additional information see 8.2.5 and 8.2.6. Smaller IF bandwidth and averaging reduce
the effect.
5.3.2.2 Noise contributions to the measurement uncertainty are in principle already included in
the characterization of the connection repeatability, see 5.3.7. It is not possible to sep-
arate the two effects and one might be tempted to neglect an extra noise contribution
to the measurement uncertainty. Neglecting an extra noise contribution is principally
possible but care should be taken, because the noise contribution varies, depending
on the VNA settings and also depending on the S-parameters of the DUT.
5.3.3.1 The VNA is being modeled as a linear network, i.e. there is a linear relation between
the forward and backward propagating waves and their detection by the receivers. The
term VNA non-linearity, sometimes also called dynamic accuracy, denotes deviations
from this behavior. The effect is dependent on the setting of the source power level
5.3.5.1 Isolation is represented by error coefficients in VNA error models, e.g. E03 and E30 in
figure D.5. It is generally of minor importance in coaxial measurements with modern
VNAs. For older VNAs it is known that it affects measurements of large attenuation
to the extent that a correction needs to be applied. Because the effect might as well
depend on the setup and has a potential to drift it should be characterized with each
measurement, as it is done for the other error coefficients.
5.3.5.2 The treatment of isolation, as done in appendices D and F, might be inadequate for
measurements other than coaxial. E.g., for on-wafer measurements additional leakage
and cross-talk paths need to be considered. This is not further discussed here. More
information can be found in [21, 22].
5.3.5.3 The procedure of characterization follows the characterization of the noise floor, as
described in G.1. The uncertainty associated with the isolation terms cannot be sepa-
rated from the noise floor. It is therefore usually justified to assume that the uncertainty
is already included in the noise floor.
5.3.5.4 If the characterization G.1 results in a mean of S21 or S12 that is significantly different
from 0, e.g. if the difference is larger than the standard uncertainty associated with the
repeated measurements of S21 or S12 , the measurement of the DUT needs to be error
corrected using the mean value of S21 or S12 as the error coefficient related to isolation.
5.3.6.1 Test port cables are sensitive to temperature changes, movement and other mechan-
ical influences. This section addresses the sometimes unavoidable movement of test
port cables. Temperature effects of cables are summarized under drift effects, as dis-
cussed in 5.3.4. Other influences inducing mechanical stress should be avoided as far
as possible, see 8.5.
5.3.6.2 Moving the test port cable(s) during calibration or DUT measurement will change the
error coefficients of the VNA. Thus, cable movement and cable torsion should be
avoided as much as possible. Related best practice is discussed in 8.5.1. For measure-
ments that involve more than one VNA port, cable movement is inevitable. The effects
of cable movement are strongly dependent on the quality of the test port cable. Even
for cables of the same type differences in stability might be rather large. Furthermore,
there is a dependency on the measurement setup and how the cables are connected to
the VNA, see 8.5. Therefore it is recommended to individually characterize the cables
in the specific measurement setup.
5.3.7.1 Deviations from ideal connector geometry lead to mechanical stress and unwanted
deformations when mating connectors. This causes changes in electrical behavior for
6 Measurement model
6.1 A VNA measurement model is the basis for the evaluation of the measurement un-
certainty. The measurement model is an analytical expression that relates inputs and
influence quantities to the final output quantities, as discussed in more detail in C.3. In
VNA metrology it relates the influence quantities discussed in 5.3 to the S-parameters
of the DUT. Using methods of uncertainty propagation, as discussed in section 7, the
uncertainties associated with the influence quantities can be propagated to evaluate
the uncertainty associated with the final result.
6.2 A measurement model is not necessarily a single large equation. It is often more
convenient and more natural to have a set of connected equations, each of which
representing either a different part of the measurement process, e.g. VNA calibration
and VNA error correction, or a refined model of one of the influences, e.g. a drift model
taking time spans between measurements and/or temperature readings into account.
6.3 A measurement model of the VNA is closely related to the error model that is applied
during VNA calibration. See appendix F for details.
7 Uncertainty evaluation
Two methods of uncertainty evaluation are presented below, the rigorous method with
uncertainty propagation through a full measurement model and the Ripple Method. It
should be pointed out that both methods deliver the same measurement result (i.e., the
same estimates of S-parameters of the DUT), when the same standards and calibration
scheme are used for VNA calibration. The different methods will, however, obtain
different measurement uncertainties for the same result. For general remarks on S-
parameter uncertainties refer to appendix C. The result of any uncertainty evaluation
should be verified regardless of the applied method, see section 4 and appendix E.
7.1.2 Rigorous uncertainty evaluation is based on a measurement model, which covers the
entire measurement process, i.e. the calibration of the VNA using calibration stan-
dards and subsequent error correction of a measurement of the DUT. Details on the
measurement model are provided in section 6 and appendix F. An essential part is
the characterization of the basic influence quantities. Following the instructions in 5.3
measurement uncertainties are assigned to the basic influence quantities. These un-
certainties are propagated through the measurement model by linear or numerical (e.g.
using the Monte Carlo Method in [5]) uncertainty propagation.
7.1.3 Based on the discussion in C.4 it might be more economical to use linear uncertainty
propagation instead of numerical methods. However, the equations of linear uncer-
tainty propagation, taking correlations fully into account, are elaborate, see [24]. It is
beyond the scope of this guide to write these equations down, because the evaluation
by hand or in a spreadsheet is not feasible. Instead software support is needed.
7.1.4 Suitable software solutions, which are able to handle the uncertainty propagation of
complex-valued quantities, are available [25, 26, 27]. These tools provide general
frameworks to realize custom-built implementations of rigorous S-parameter uncer-
tainty evaluation. For software solutions that are specifically targeting S-parameter
measurements see [28, 19, 29, 30]. These solutions already contain the VNA mea-
surement models and support different calibration algorithms, i.e. programming is not
necessary. Questions related to validation of a software are not addressed in this
guide, because they can’t be generally answered and have to be evaluated case by
case.
7.1.5 Depending on the implementation of rigorous uncertainty propagation it not only pro-
vides the correlation between the two components of a complex-valued S-parameter,
but also the correlation information between the individual S-parameters and possibly
even the cross-frequency correlations. This is valuable if the measured S-parameters
are used to calculate derived quantities.
7.1.6 It is a disadvantage of the rigorous method that the equations of the measurement
model can’t be easily printed in this guide and the user needs to rely on external soft-
ware. All other elements that are needed to implement the method are available in this
guide.
7.1.7 The traceability to SI units is established through the calibration standards. It is there-
fore necessary to have an SI traceably characterized calibration kit to implement the
rigorous method. See the remarks in section 2 as well.
7.2.7 An advantage of the Ripple Method is that everything needed to apply the method is
documented in this guide and it is not necessary to rely on external specialized software
to implement it. The equations in appendix H are sufficiently simple to be implemented
in a spreadsheet.
7.2.8 It has been demonstrated that it is possible to increase the quality of the Ripple Method
with considerable additional effort. Many of the shortcomings mentioned in 7.2.3 can
be resolved this way. The enhanced method requires a characterized air-dielectric line
with a positionally controlled center conductor and advanced methods of data process-
ing. Due to the complexity of the method it is not considered in this guide. Though
it should be pointed out that in this improved form the Ripple Method can be a valid
alternative to evaluate the measurement uncertainty of S-parameter measurements.
For details see [33, 34, 35, 36, 37, 38].
8.1.2 Changes in the temperature during and after calibration might lead to drift in the er-
ror coefficients of the VNA and will directly affect measurements of the DUT. This is
generally more significant than the effect of absolute deviation from standard labora-
tory temperature on the calibration standards. Drift occurring after calibration can be
specified for specific environmental conditions, see 5.3.4, and needs to be considered
in the evaluation of the measurement uncertainty. A measurement setup including test
port cables is more sensitive to temperature effects. On the other hand a test port
cable decouples the measurement devices from the VNA test port. Heat flow from the
VNA test port might otherwise lead to heating of the components and change their
electrical characteristics. For reasonably stable environmental conditions it is therefore
recommended to use a test port cable.
8.2.6 Noise can be reduced through smoothing as well. This functionality should be used
with care, however. Pronounced features in the frequency dependence, e.g. in a mea-
surement of a filter, might be altered. Smoothing has the potential to introduce addi-
tional measurement error if used negligently.
8.3.2 Many calibration kits contain sliding loads to replace the fixed load at higher frequen-
cies. A sliding load increases the accuracy of the measurement if the physically im-
perfect fixed load is assumed to be ideal, i.e. without reflection. The assumption of an
ideal load is usually made in combination with the polynomial representation of open
and short. A sliding load is not needed if the definition of the fixed load is data-based.
See comments in 2.2.7 and 2.2.8.
8.3.3 A key quality characteristic of mechanical standards is stability, which has a direct
impact on the accuracy of measurement. A high connection repeatability with low
sensitivity to connector orientation is essential (see 5.3.7 and 8.8), as is long term
stability (see 5.3.1.5).
8.3.4 Quality criteria for mechanical calibration standards that can be assessed visually are:
concentricity of center conductor, integrity of contact fingers of female interfaces and
surface finish of the electrical contact areas. The electrical stability of the standards is
directly impacted by these factors.
8.3.5 Some mechanical calibration and verification standards or test port adapters are avail-
able in two different designs of the female connector interface, slotted and slotless.
Slotted interfaces are more sensitive to variations of the male pin diameter compared
to an optimized slotless design. Slotless designs, on the other hand, might have a
larger reflection coefficient. From a metrological point of view the latter is of less con-
cern. A slotless design is therefore preferable, because it generally provides better
repeatability. See the notes in 8.5.2 as well.
8.3.6 Calibration standards can change their characteristics because of heat flow caused by
mating with a warm test port or by touching, see 8.1. Both should be avoided as much
as possible. Typical indicators of this problem are phase changes of highly reflective
8.4.2 The impedance states of an ECU are subject to aging and hence the recalibration
intervals have to be chosen accordingly. Long-term stability tests under controlled
laboratory conditions are indicating that ECU states are stable on the time scale of
months and years. However, this is very much dependent on the design of the ECU
and on the operating conditions. It is therefore impossible to generally quantify the
recalibration interval. It is advised to start with short recalibration intervals to build an
evidence-based history of the ECU and then adjust the interval accordingly.
8.5.1.1 An appropriate fixture that supports and defines the layout of the test port cables is a
necessary accessory for accurate VNA measurements. A simple cable fixture can be
realized with a plain and solid surface in front of the VNA (e.g. a commercially available
breadboard) using foam pads and clamps. An example is shown in figure 8.1. The
fixture should keep the test port cable fixed during one-port measurements. For two-
port measurements one cable remains fixed and the other one should be arranged so
that unnecessary cable movements are avoided. If possible, the test port cables should
be maintained in a horizontal plane defined by the test port connectors of the VNA.
Cable twisting, extreme curvature and stress on the connection should be avoided.
8.5.1.2 High performance test port cables often have a natural bend due to the production pro-
cess. Maintaining the natural bending helps to reduce measurement error due to cable
movement. Even then it is still possible that small angular changes in the connection
of the test port cable to the VNA can lead to differences in phase stability. This can
be examined and optimized by connecting a short at the reference plane (end of the
test port cable) and observing how phase variations depend on cable movement while
changing slowly the angular orientation of the connection between test port cable and
VNA.
8.5.1.3 For reasons given in 8.5.2 it is recommended to use an adapter in combination with
the test port cable. The most stable cable-adapter combination is achieved by using
ruggedized connectors at the cable-adapter interface. The ruggedized connector can
be found on many VNA test ports. It has a large threaded body that helps to stabilize
the connection.
8.5.1.4 The selection of the correct VNA calibration scheme can help to reduce measurement
error due to cable movement as well, as discussed in 3.2.
8.5.2.1 It is recommended to use test port adapters for several reasons. If any damage occurs
to the connector only the adapter has to be changed or repaired, rather than the entire
test port cable or VNA test port. Using a metrology grade adapter with geometries
close to the nominal values and good concentricity of the center conductor will minimize
mechanical stress and improve electrical repeatability. Finally, an optimal pin depth with
respect to the reference plane, see 8.5.2.2, can be established on the test port side.
8.5.2.2 In the past, the objective was always to have a connection at the measurement refer-
ence plane as flush as possible, i.e. the pin gap at the center conductor was kept as
small as possible. It has been shown [23] that this causes coupling effects, resulting in
an undefined reference plane, and possibly even resonance effects and repeatability
8.5.2.3 The simple stability test described in 8.7 is a good method to identify any stability
problems related to the test port adapters, i.e. a loose connection between test port
cable and adapter.
8.6.2 Metrology grade connectors are precision components that require special care. Over-
torquing, rotating the body of the component during connection and rough abrasive
handling should be avoided. Cleaning, visual inspections, and pin depth measure-
ments should be carried out regularly. Dirty connectors often show degraded repeata-
bility. This especially holds for small connectors. It is not only the area of the contact
zones that is sensitive to contamination (resulting in magnitude and phase changes).
Even a contaminated thread of the connecting nut might affect the impact of the torque
force and degrade the repeatability. Protruding center pins and defects might cause
damage when mated with other components. Mechanical stress on the connection
should be avoided, see 8.5.1, because it will degrade the electrical performance of the
measurement or even lead to damage.
8.8 Repeatability
8.8.1 It is recommended to repeat each measurement for at least four different connector
orientations. The purpose of this practice is partly to verify the specific repeatability
of a certain connector type, but also to see if any component has a stability issue. It
is not uncommon that ill-behaving components show two or more different electrical
1.0 mm 5 2
1.85 mm 5 2
2.4 mm 15 6
2.92 mm 10 4
3.5 mm 15 6
Type-N (50 Ω) 12 5
states imposed by the varying mechanical stress during the measurements at different
connector orientations.
8.8.2 It is important to completely disconnect the center conductor before changing the con-
nector orientation. This prevents rotational stress on the female contact fingers and
the male pin. Rotational stress might bend the contact fingers, resulting in a change
of contact point, or damage the surface plating. The operator must exercise care. Bad
alignment of the connectors, excessive pushing or pulling during mating and discon-
necting will provide extra stress on the center conductor, impacting structures or circuits
attached to it, as e.g. soldering joints or resistive elements.
[3] M. Hiebel. Fundamentals of Vector Network Analysis. Rohde & Schwarz, 2007. ISBN
978-3-939837.
[4] Keysight Technologies. Network Analyzer Basics. Electronic document. Back to Basics
Series, available at www.keysight.com.
[5] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of measurement
data - Supplement 2 to the "Guide to the expression of uncertainty in measurement"
- Extension to any number of output quantities, 2011. JCGM 102:2011; available at
http://www.bipm.org/en/publications/guides/.
[6] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of Measurement Data -
Guide to the expression of uncertainty in measurement, 2008. JCGM 100:2008; available
at www.bipm.org/en/publications/guides/gum.html.
[7] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. International Vocabulary of Metrol-
ogy - Basic and General Concepts and Associated Terms (VIM 3rd edition), 2012. JCGM
200:2012; available at http://www.bipm.org/en/publications/guides/.
[8] R. B. Marks and D. F. Williams. A General Waveguide Circuit Theory. J. Res. Natl. Stand.
Technol., 97(5):533 – 562, 1992.
[11] N. M. Ridler and A. G. Morgan. New primary reference standard for vector network anal-
yser calibration at millimetre wavelengths in coaxial line. Meas. Science and Technol.,
19:065103, 2008.
[13] K. Wong and J. Hoffmann. Improve VNA Measurement Accuracy by Including Connector
Effects in the Models of Calibration Standards. In ARFTG Conference Digest, number 82,
pages 1–7, 2013.
[16] P. I. Somlo and J. D. Hunter. Microwave Impedance Measurements - IEE electrical mea-
surement series; 2. Peter Peregrinus Ltd, London, UK, 1985.
[17] D. M. Pozar. Microwave Engineering. John Wiley & Sons, Inc., 4th edition, 2011. ISBN
978-0-470-63155-3.
[18] A. Ferrero and U. Pisani. Two-port network analyzer calibration using an unknown thru.
IEEE Microwave & Guided Wave Letters, 2(12):505 – 507, 1992.
[20] K. Wong. Receiver Linearity Measurement Using Two CW Signals. In ARFTG Conference
Digest, number 78, pages 1–5, 2011.
[22] A. Ferrero and F. Sanpietro. A Simplified Algorithm for Leaky Network Analyzer Calibration.
IEEE Microwave & Guided Wave Letters, 5(4):119 – 121, 1995.
[23] J. Hoffmann, P. Leuchtmann, and R. Vahldieck. Pin Gap Investigations for the 1.85 mm
Coaxial Connector. In Proceedings of the 37th European Microwave Conference, num-
ber 76, pages 388 – 391, 2007.
[24] M. Garelli and A. Ferrero. A Unified Theory for S-Parameter Uncertainty Evaluation. IEEE
Trans. Microwave Theory & Tech., 60(12):3844 – 3855, 2012.
[25] B. D. Hall. Computing uncertainty with uncertain numbers. Metrologia, 43:L56 – L61,
2006.
[27] B. D. Hall. Object-oriented software for evaluating measurement uncertainty. Meas. Sci.
Technol., 24:055004, 2013.
[30] Dynamic uncertainty for S-parameters, Option 015, Keysight Technologies, available
through www.keysight.com.
[33] F. Mubarak and G. Rietveld. Uncertainty Evaluation of Calibrated Vector Network Analyz-
ers. IEEE Trans. Microwave Theory & Tech., PP(99):1 – 13, 2017.
[37] J. Stenarson and K. Yhland. Residual error models for the SOLT and SOLR VNA calibra-
tion algorithms. In ARFTG Conference Digest, number 69, pages 133 – 139, 2007.
[38] J. Stenarson and K. Yhland. A new assessment method for the residual errors in SOLT
and SOLR calibrated VNAs. In ARFTG Conference Digest, number 69, pages 140 – 145,
2007.
[40] J. Hoffmann and P. Huerlimann. Key parameters of coaxial connector models - mechan-
ical design features and electrical properties. electronic document, 2015. available at
www.metas.ch/hf/docs.
[42] J. Hoffmann and J. Ruefeneacht. How to make good coaxial connections. electronic
document, 2013. available at www.metas.ch/hf/docs.
[44] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML. Evaluation of measurement data
- Supplement 1 to the "Guide to the expression of uncertainty in measurement" - Propa-
gation of distributions using a Monte Carlo method, 2008. JCGM 101:2008; available at
http://www.bipm.org/en/publications/guides/.
[46] A. Ferrero and U. Pisani. QSOLT: a new fast calibration algorithm for two port S-parameter
measurements. In ARFTG Conference Digest, number 38, pages 15 – 24, 1991.
[47] O. Ostwald. T-Check accuracy test for vector network analyzers utilizing a Tee-junction.
Electronic document, June 1998. Rohde & Schwarz Application Note 1EZ43_0E, Available
at https://www.rohde-schwarz.com.
[51] R. Willink and B. D. Hall. A classical method for uncertainty analysis with multidimensional
data. Metrologia, 39:361 – 369, 2002.
[53] B. D. Hall. Expanded uncertainty regions for complex quantities. Metrologia, 50:490 –
498, 2013.
[54] B. D. Hall. Expanded uncertainty regions for complex quantities in polar coordinates.
Metrologia, 52:486 – 495, 2015.
[55] R. B. Marks. Formulations of the basic vector network analyzer error model including
switch-terms. In ARFTG Conference Digest, number 50, pages 115 – 126, 1997.
[56] D. Blackham. Application of weighted least squares to OSL vector error correction. In
ARFTG Conference Digest, number 61, pages 11–21, 2003.
[59] M. Wollensack, J. Hoffmann, J. Ruefenacht, and M. Zeier. VNA Tools II: S-parameter
uncertainty calculation. In ARFTG Conference Digest, number 79, pages 1 – 5, 2012.
[65] IEEE. IEEE 287-2007 Standard for Precision Coaxial Connectors (DC to 110 GHz), 2007.
[66] T. Reichel, R. H. Judaschke, and F. Rausche. Effect of Uncorrected Test Port Match of a
Vector Network Analyzer on the Transmission Coefficient Gained after SOLT Calibration.
accepted by CPEM conference, 2018.
[67] N. M. Ridler and C. Graham. Some typical values for the residual error terms of a calibrated
vector automatic network analyser (ANA). In Proc BEMC Conference, pages 45/1 – 45/4,
1999. Brighton, UK.
[68] J. Hoffmann. Errors in the Ripple Technique due to Pin Gap. electronic document, 2016.
available at www.metas.ch/hf/docs.
[69] B. D. Hall. On the expression of measurement uncertainty for complex quantities with
unknown phase. Metrologia, 48:324 – 332, 2011.
[70] T. Reichel. Derivation of the Uncertainty assigned to the Residual Reflection Track-
ing Parameter with SOL(T) calibrations. electronic document, 2018. available at
www.metas.ch/hf/docs.
[71] A. R. Kerr. Mismatch caused by waveguide tolerances, corner radii, and flange
misalignment. Technical report, National Radio Astronomy Observatory, Char-
lottesville, VA, USA, 2009. Electronics Division Technical Note No 215, available at
http://www.gb.nrao.edu/electronics/edtn/edtn215.pdf.
with u(Re [x̂]) and u(Im[x̂]) the standard uncertainties associated with the real and
imaginary components and r(Re[x̂], Im[x̂]) the correlation coefficient between the real
and imaginary component estimates.
C.2.1 Correlation
Correlation in measured values is caused by common influences. A few examples: the
use of the same reference standard for the characterization of calculable calibration
standards will cause correlations between the mechanical measurements of calculable
standards. If these standards are used to characterize transfer standards, i.e. an SOL
kit, short, open and matched load will be correlated. The measurements of a short stan-
dard during a VNA one-port calibration will influence the determination of the reflection
tracking and the source match resulting in correlation between them. The multiplication
of measured values for two complex-valued quantities (i.e. two reflection coefficients)
in Cartesian coordinates will lead to correlation between the real and imaginary com-
ponents of the product, even if the initial measurements were uncorrelated (i.e. the
quantity estimates are not correlated).
Correlation in VNA measurements is unavoidable, and it can either increase or
decrease the measurement uncertainty. The best way to take correlation effects into
account is through refined modeling of the measurement process. Common influence
quantities are part of the measurement model. Rigorous uncertainty propagation based
on such a measurement model, see 7.1, will take correlation properly into account. For
random effects it is possible to evaluate correlation in a statistical way by using the
formalism in C.5.
The terms that appear in a measurement model (C.3) refer to actual quantities. But
as pointed out in C.2 there are only estimates available. An estimate of the quantity
with the scalar variance v(·) and the scalar mean m(·) according to equations (C.5) and
(C.4), respectively. The covariance matrix of the mean is accordingly
1
v(m(x)) = v(x) . (C.11)
n
The formalism for complex-valued quantities is defined in direct analogy to the scalar
case with the notable addition of the correlation information, which is contained in v21 in
equation (C.10). The same principle formalism holds for vectors of length N , resulting
in a mean vector of length N for scalar quantities (2N for complex-valued quantities)
and covariance matrix of size N × N for scalar quantities (2N × 2N for complex-valued
quantities).
with the standard deviation of the mean s(m(x)) according to equation (C.8).
For a complex-valued quantity x the mean according to equation (C.9) is the best
estimate x̂ with an uncertainty matrix u(x̂)
with the covariance matrix of the mean v(m(x)) according to equation (C.11).
C.6.1 Method according to GUM Supplement 2 [5]: if the Type A uncertainties are calculated
with the equations (C.14) and (C.15), the coverage factors for the combined uncertainty
are k = 1.96 for 95% coverage in the scalar case and k = 2.45 for 95% coverage in
the two-dimensional case. The latter can be applied to a complex-valued quantity and
corresponds to an elliptical region in the complex plane.
C.6.2 If the Type A uncertainties are calculated according to equations (C.12) and (C.13) and
propagated to the combined uncertainty, it is necessary to calculate effective degrees
of freedom by applying the Welch-Satterthwaite formula. The coverage factor is then
dependent on the effective degrees of freedom. This is described in the ISO-GUM
[6] for the case of uncorrelated scalar quantities. A generalization of the procedure to
multivariate and complex-valued quantities is described with an example in [50]. Further
details are given in [51, 52].
C.6.3 The discrepancies between the two methods described in C.6.1 and C.6.2 become
apparent when Type A uncertainties based on a small sample sizes (as discussed in
C.5.3.2) make a significant contribution to the overall uncertainty. Otherwise the two
approaches converge to the same coverage factors. The method in C.6.1 is easier to
apply, but there are indications [52] that it provides overly conservative coverage regions
compared to the method in C.6.2.
which is used to perform the error correction on the DUT measurement. The previ-
ously determined estimates of the error coefficients are substituted into equation (D.2)
together with Sm11 for the DUT to obtain an estimate of the S-parameter, Ŝ11 , for the DUT.
Ŝ11 is sometimes also referred to as the error corrected S-parameter.
The most common one-port calibration method for coaxial measurements is based
on a short, an open and a matched load as calibration standards. Common acronyms
are SOL (short - open - load) or OSM (open - short - match).
An alternative to SOL is to use a set of shorts with different offset lengths. However,
this technique is band-limited and more than three offset shorts are necessary to cover
larger frequency ranges. The acronym for this technique is SSS. It is based on calcu-
lable standards and can therefore be used for primary experiments, see the note on
E00 Directivity
E01 Reflection tracking
E11 Source match
E00 E11
S11
E01
b1
Figure D.1: Signal flow graph representing a one port VNA error model with the error
coefficients listed in table D.1, which relate indicated reflection coefficient Sm
11 = b1 /a1
with true reflection coefficient S11 .
SSST in D.2.2. An advantage of SSS is that shorts are typically more robust and stable
than other types of standards. A disadvantage is the bandwidth limitation. It is possible
to overcome the bandwidth limitation at low frequencies by using additional open and
matched load standards.
• In the first state, the test port is connected to the source and two receivers mea-
sure the incoming and outgoing waves. This corresponds to the configuration of
one-port measurements and is characterized by the three error coefficients, di-
rectivity, source match and reflection tracking, which can be determined with a set
of three one-port standards, as discussed in D.1.
• In the second state, the port is connected to the switch termination, and incoming
waves are measured by only one receiver. This state is modeled with two error
a1 a2
b1 b2 b1 b2
coefficients, the load match and the transmission tracking, which can be deter-
mined by the measurement of a two-port standard. With a four-receiver VNA, the
corresponding reference channel is not used.
Figure D.2 (right side) shows port 1 in the first state and port 2 in the second state.
Changing the switch will put port 2 into the first state and port 1 into the second state.
This results in two error models. The forward model (figure D.3), relates indicated and
true forward S-parameters, Sm m
11 , S21 and S11 , S21 , respectively. The reverse model
(figure D.4), relates indicated and true reverse S-parameters, Sm m
22 , S12 and S22 , S12 ,
respectively. The error coefficients are listed in table D.2. The ten-term error model
is also referred to as twelve-term error model, if isolation terms are also taken into
account.
Table D.2: List of error coefficients of forward and reverse ten-term error model.
EF00 ER
33 Directivity
EF01 ER
32 Reflection tracking
EF11 ER
22 Source match
EF22 ER
11 Load match
EF32 ER
01 Transmission tracking
EF30 ER
03 Isolation
1 S21 EF32
a0 b2
EF01 S12
b1
Figure D.3: Signal flow graph representing the ten-term forward VNA error model (in-
cluding the isolation term) for two port measurements with the error coefficients listed in
table D.2. The model relates indicated forward S-parameters, Sm m
11 = b1 /a0 and S21 = b2 /a0 ,
with true S-parameters, S11 and S21 .
S21 ER
32
b2
ER
11 S11 S22 ER
22 ER
33
ER
01 S12 1
b1 a0
ER
03
Figure D.4: Signal flow graph representing the ten-term reverse VNA error model (includ-
ing the isolation term) for two-port measurements with the error coefficients listed in
table D.2. The model relates indicated reverse S-parameters, Sm m
22 = b2 /a0 and S12 = b1 /a0 ,
with true S-parameters, S22 and S12 .
a1 1 S21 E32 b2
Figure D.5: Signal flow graph representing the seven-term VNA error model (including
isolation terms) for two port measurements with the error coefficients and the switch
terms listed in table D.3. The model relates indicated S-parameters, Sm m
11 = b1 /a1 , S21 =
m m
b2 /a1 , S22 = b2 /a2 and S12 = b1 /a2 , with true S-parameters, S11 , S21 , S22 and S12 .
This equation relates the error corrected reflection coefficient Sc11 to the true reflection
coefficient S11 of the DUT via the residual one-port error coefficients, δE00 (directivity),
δE11 (source match) and δE01 (reflection tracking). Since the sensitivity of Sc11 to the
individual residual error coefficients depends on the reflectivity of the DUT, S11 , equa-
tion (E.1) might serve as a guideline to choose verification standards appropriately to
identify potential sources of erroneous calibration.
These equations relate the error corrected S-parameters, Sc11 and Sc21 , to the true S-
parameters S11 and S21 of the DUT via the residual error coefficients, δEF00 (directivity),
δEF11 (source match), δEF01 (reflection tracking), δEF22 (load match) and δEF32 (transmis-
sion tracking). Analogous equations exist for the reverse residual error model
E.1.2.5 T-checker
An quick and easy test of VNA calibration validity utilizes a common T-junction, one
port of which is terminated by a load impedance, Z [47] as shown in figure E.3. The
50 Ω 25 Ω 50 Ω
−10 −0.5
−20 −1
|S11 | / dB
|S21 | / dB
−30 −1.5
−40 −2
−50 −2.5
|S11 | |S21 |
−60 −3
0 5 10 15
Frequency /GHz
Figure E.2: Simulated S-parameter magnitudes of an ideal Type-N Beatty line with a 25 Ω
section length of 70 mm.
1 2
with ∗ denoting the conjugate transpose. The S-parameters in the equation refer to
measurements at the ports labelled 1 and 2 in figure E.3. If the T-junction is lossless,
cT is unity, independent of the load impedance Z, which can be chosen arbitrarily.
If the lossless condition is fulfilled, which might be approximately the case at low
frequencies (< 6 GHz), deviation of cT from unity indicates measurement inaccuracies
of the VNA under investigation. As a rule of thumb, deviations in the range 1 ± 0.1
can be regarded as minor. Since the T-check is sensitive to both one- and two-port
error coefficients, it enables an overall verification; however, it might not help to identify
inaccuracies of individual error coefficients.
For a truly quantitative statement based on cT = 1 the losslessness of the T-junction
needs to be demonstrated and uncertainties associated with the experimental and theo-
retical parameter cT need to be determined. Unless this is done the T-checker provides
merely a qualitative verification test.
A characterized T-checker can be used for quantitative verification as can any other
stable two-port device. Reflection and transmission coefficients of commercially avail-
able T-checker devices are typically in the order of 0.2 to 0.4 and -2 dB to -5 dB, respec-
tively.
E.1.2.6 Flush thru
After an SOLR calibration it is most useful to measure a flush thru by directly connecting
the two reference planes, provided that the DUT is insertable. Ideally, this measurement
should by definition result in S11 = S22 = 0 and S21 = S12 = 1, allowing for measure-
ment uncertainties. In order not to undermine the SOLR advantage of minimizing cable
movements this measurement should be done at the end after all calibration standards
and the DUT have been measured. If the VNA firmware or the software permit an error
correction of an earlier measurement it can also be done at the beginning before all
other measurements start.
After an SOLT calibration followed by a two-port measurement of a passive, recip-
rocal DUT the following test can be carried out: the measurement data of the DUT
can be used to perform an SOLR calibration. The SOLR error correction can then be
applied to the raw measurement of the flush thru, which has already been measured
during the SOLT calibration. The result should again provide the ideal values, allowing
for measurement uncertainties.
The verification measurement of a flush thru is a highly quantitative test that helps to
identify problems related to the characterization data for the highly reflective standards
that were used during VNA calibration.
The quantity d is the frequency dependent difference between measurement and refer-
ence of a scalar quantity related to the verification standard. The scalar quantity is e.g.
the magnitude of S11 . An estimate of d is in this case
ˆ
d = Ŝc11 − Ŝr11 ,
(E.6)
with the error corrected measurement and the reference data denoted by the super-
scripts c and r, respectively.
ˆ is the standard uncertainty associated with d.
u(d) ˆ It is calculated from the individ-
c r
ual uncertainties associated with |Ŝ11 | and |Ŝ11 | and, if available, from the associated
correlation coefficient r(|Ŝc11 |, |Ŝr11 |):
r 2 2
u dˆ = u Ŝc11 + u Ŝr11 − 2r Ŝc11 , Ŝr11 u Ŝc11 u Ŝr11 . (E.7)
k is the coverage factor used to expand the uncertainty in the denominator of equa-
tion (E.5). The value of k determines the normalization factor and determines the sta-
tistical significance of the pass criteria. Larger k represent more relaxed conditions. A
value that is often used is k = 1.96. It expands the uncertainty associated with dˆ to a
95% coverage interval. k = 1 corresponds to a 68% coverage interval. This verification
procedure assumes that the underlying PDF associated with dˆ is Gaussian.
Equations (E.6) and (E.7) are shown for the magnitude of S11 . Analog expressions
can be used for the phase of S11 or any other scalar component of an S-parameter. In
case of the phase of the S-parameter it is advised to use the unwrapped phase to avoid
problems related to the cyclicality of the phase.
with superscripts c and r denoting error corrected measurement and reference, respec-
tively.
u(d̂) is the uncertainty (covariance) matrix associated with d̂. It is expressed as
u d̂ = u Ŝc11 + u Ŝr11 − u Ŝc11 , Ŝr11 − u Ŝr11 , Ŝc11 (E.10)
with u(Ŝc11 ) and u(Ŝr11 ) according to equations (C.1) and (C.2) and
h i h i h i h i
u Re Ŝ c , Re Ŝr u Re Ŝ c , Im Ŝr
11 11 11
u Ŝc11 , Ŝr11 = h i h i h i h 11 i (E.11)
u Im Ŝc11 , Re Ŝr11 u Im Ŝc11 , Im Ŝr11
h i h i h i h i
u Re Ŝr , Re Ŝc u Re Ŝ r , Im Ŝc
11 11 11
u Ŝr11 , Ŝc11 = h i h i h i h 11 i . (E.12)
u Im Ŝr11 , Re Ŝc11 u Im Ŝr11 , Im Ŝc11
The elements in the matrices represent covariances, which are generally expressed as
u(â1 , â2 ) = r(â1 , â2 ) u(â1 ) u(â2 ) with r(â1 , â2 ) denoting the correlation coefficient and
u(â1 ) and u(â2 ) denoting standard uncertainties associated with estimates of a1 and
a2 , respectively. (E.11) and (E.12) are zero unless there is correlation between Ŝc11 and
Ŝr11 .
The coverage factor k plays the same role as in the scalar case, but for 95% cover-
age probability in the two-dimensional case the value k = 2.45 should be used, see [5].
For k = 1 the coverage probability is 39%. Again, this assumes that the probability den-
sity function associated with the measurement uncertainty of d̂ is a bivariate Gaussian
one.
Equations (E.9), (E.10), (E.11) and (E.12) are valid for S11 . Analog expressions can
be used for the other S-parameters. Instead of using real and imaginary components
the same formalism can be applied for magnitude and phase, but this should only be
done in cases where the phase is well defined, see remarks in C.1.
The same formalism can be applied to dimensions larger than two, by e.g. using all
four S-parameters that have been measured for a two-port device. This corresponds to
an eight-dimensional evaluation and the factor k needs to be adjusted accordingly.
0.3
0.25
0.2
4
k=1
ǫ (|S11 |)
k = 1.96
2
0.1
Reference data
Measurement data
0.05
arg (S11 )
−0.05
−0.1
4
ǫ (arg (S11 ))
k=1
k = 1.96
2
4
k=1
ǫ (S11 )
k = 2.45
2
0
10 20 30 40 50 60
Frequency /GHz
Figure E.4: Quantitative verification criteria applied to the verification measurement of a
T-checker. Detailed explanation can be found in the text.
EURAMET Calibration Guide No. 12
Version 3.0 (03/2018) – 65 –
0.03
Reference data
Measurement data
0.02
0.01
Re [S11 ]
−0.01
−0.02
4
ǫ (Re [S11 ])
k=1
k = 1.96
2
0.02
0.01
0
Im [S11 ]
−0.01
k=1
k = 1.96
2
4
k=1
ǫ (S11 )
k = 2.45
2
0
10 20 30 40 50 60
Frequency /GHz
Figure E.5: Quantitative verification criteria applied to the verification measurement of a
matched load. Detailed explanation can be found in the text.
EURAMET Calibration Guide No. 12
Version 3.0 (03/2018) – 66 –
F VNA measurement models
F.1 Introduction
The propagation of input measurement uncertainties to the final result is based on VNA
measurement models. Such models are not unique, i.e. different levels of refinement
are possible. A VNA measurement model incorporates the same error model used
during the calibration of the VNA, see section 3, and adds the additional quantities
discussed in 5.3, which are influencing the measurements.
Sections F.2, F.3 and F.4 discuss rigorous models, covering the entire VNA mea-
surement process. Section F.5 introduces a simplified residual model, which is used by
the Ripple Method. Section F.6 briefly discusses the influence quantities in the models
with further references to the characterization of associated uncertainties in appendix
G.
Equation (F.2) can be used to determine an estimate of the reflection coefficient of the
DUT by substituting the error coefficients determined in the calibration process.
E00 + D00
E11 + D11 C00 C11 S11
NL
Figure F.1: Signal flow graph of one-port VNA measurement model. The model repre-
sents an extension of the one-port VNA error model with additional terms influencing the
measurements.
Table F.1: List of additional influence quantities in the one-port measurement model
shown in figure F.1.
Symbol Description
NL Noise floor
NH Trace noise
L Non-linearity
D00 Drift of directivity
D01 Drift of reflection tracking
D11 Drift of source match
C00 C11 Reflection of cable and connector
C01 C10 Transmission of cable and connector
1 1 C10 S21
a0
EF00 + DF00
EF11 + DF11 C00 C11 S11 S22
NH1 · L1 EF01 · DF01 C01 S12
b1
Figure F.2: Signal flow graph of two-port VNA measurement model. The model repre-
sents an extension of the ten-term forward VNA error model with additional terms influ-
encing the measurements. Due to the limitations in page width the graph is separated in
two parts, connected at the dashed lines. Accordingly, the measurement model for the
reverse mode is identically based on the ten-term reverse VNA error model.
E30
1 1 C10 S21
a1
E00 + D00
W00 + V00 E11 + D11 C00 C11 S11
Figure F.3: Signal flow graph of two-port VNA measurement model (only the port 1 side
is shown). Port 2 is symmetrical with the dashed line, except that both reflection track-
ing terms have to be considered at the port 2 side, see table D.3 and figure D.5. The
model represents an extension of the seven-term VNA error model with additional terms
influencing the measurements.
δE00 + D00
Sc11 δE11 + D11 C00 C11 S11
NL
Figure F.4: Signal flow graph of residual one-port VNA measurement model. The model
is equivalent to the one-port measurement model shown in figure F.1, except that the
error coefficients Eij are replaced by residual error coefficients δEij and the indicated
reflection coefficient Sm c
11 is replaced by the error corrected reflection coefficient S11 .
Using the same approach as for the one-port case, residual two-port measurement
models can be formulated, e.g. the corresponding forward model would be based on
figure F.2.
The residual measurement model has the advantage that it leads to relatively simple
equations for the propagation of measurement uncertainties. The Ripple Method, which
is discussed in appendix H, is based on the residual measurement model with some
further simplifications. The resulting equations for linear uncertainty propagation are
shown in H.4.
It is known, however, from experience that in transmission measurements the load
match has an amplifying effect on the uncertainty, which is often not negligible. Such an
effect is simply not in the scope of the residual measurement model and it is necessary
to take additional measures to avoid an underestimation of the uncertainty, see H.4.2.1.
Use of the full model, as done for the rigorous uncertainty evaluation, doesn’t have that
problem.
CA2T COR
C00 = CAR +
1 − COR CAR
CAR
C11 = COR +
1 − COR CAR
CAT
C10 = C01 = .
1 − COR CAR
These expressions simplify considerably if used for linear uncertainty propagation with
best estimates of unity and zero.
The term NLi denotes noise floor with N̂Li = 0. The terms NHi and Li denote
trace noise and non-linearity, respectively, with N̂Hi = 1 and L̂i = 1. The associated
uncertainties can be determined using the procedures in G.1 and G.2, respectively.
Cable Connector
CAT 1
CAR CAR
COR COR
CAT 1
Figure F.5: Signal flow graph representing a simple measurement sub-model for cable
movement and connector repeatability at port one.
a) VNA settings: leave the VNA uncalibrated and set the test port power, IF band-
width and averaging to the values that are being used for measurements. Set the
sweep type to CW (continuous wave).
b) Connect two shorts or opens to port 1 and to port 2. Choose the standard, which
shows a larger reflection coefficient. Wait a few minutes.
e) Define a reasonable envelope that encloses the calculated values over the whole
frequency range. Typically, the whole frequency range is divided into frequency
bands, each with a constant value that encloses the calculated values. The values
can be taken as the standard uncertainties associated with the real and imaginary
components of N̂L2 .
h) For the values defined in step f) and g), define envelopes according to step e) and
assign the values as uncertainties associated with magnitude and phase of N̂H1 .
Figures G.1, G.2, G.3, G.4, G.5 and G.6 are illustrating the above procedure.
0.5
21 ]
Re [Sm
−0.5
−1
−1.5
0 10 20 30 40 50 60 70
Frequency /GHz
Figure G.1: Characterization of VNA noise floor according to the procedure in G.1. The
blue data points are measurements of Sm 21 according to step c) of the procedure. Only the
real component of the S-parameter is shown. The imaginary component is not shown
here, but gives very similar results. 801 measurements were made per frequency point.
The black circles with bars indicate mean and standard deviation s(Re [Sm 21 ]) of the mea-
sured sample at each frequency point, according to step d). The orange line indicates a
possible envelope, according to step e). Note that the lowest frequency in this example
is 1 GHz. At lower frequencies significantly larger noise floor is conceivable, see figure
G.2.
−3
x 10
6
2
21 ]
Re [Sm
−2
−4
−6
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz
Figure G.2: The same as figure G.1, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.
1.5
1
11 /m (S11 )| − 1
0.5
0
m
−0.5
|Sm
−1
−1.5
−2
−2.5
0 10 20 30 40 50 60 70
Frequency /GHz
Figure G.3: Characterization of VNA trace noise magnitude according to the procedure
in G.1. The blue data points are normalized measurements of Sm 11 according to step c) of
the procedure. 801 measurements were made per frequency point. The black circles with
bars indicate mean and standard deviation of the displayed sample at each frequency
point, according to step f). The orange line indicates a possible envelope, according to
step h). Larger values of trace noise can occur for values at lower frequencies, see figure
G.4.
−3
x 10
2.5
1.5
1
11 /m (S11 )| − 1
0.5
m
−0.5
|Sm
−1
−1.5
−2
−2.5
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz
Figure G.4: The same as figure G.3, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.
0.03
0.02
◦
11 /m (S11 )) /
0.01
m
0
arg (Sm
−0.01
−0.02
−0.03
−0.04
0 10 20 30 40 50 60 70
Frequency /GHz
Figure G.5: Characterization of VNA trace noise phase according to the procedure in G.1.
The blue data points are normalized measurements of Sm 11 according to step c) of the
procedure. 801 measurements were made per frequency point. The black circles with
bars indicate mean and standard deviation of the displayed sample at each frequency
point, according to step g). The orange line indicates a possible envelope, according to
step h). Larger values of trace noise can occur for values at lower frequencies, see figure
G.6.
0.15
0.1
0.05
◦
11 /m (S11 )) /
m
0
arg (Sm
−0.05
−0.1
−0.15
−0.2
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Frequency /GHz
Figure G.6: The same as figure G.5, but for frequencies below 0.1 GHz. Note the sig-
nificantly enlarged noise level towards low frequencies due to the reduced efficiency of
couplers.
a) Set the test port power of the VNA to the value that is being used for the mea-
surements.
c) Perform a full two port calibration of the VNA. The SOLR calibration is preferred
in order to avoid unnecessary cable movement.
e) Measure all states of the step attenuation device over the whole frequency range
and perform a VNA error correction to obtain estimates, Ŝ21 (j) and Ŝ12 (j), of
the S-parameters of the attenuation device for each state j = 0, 1, 2, 3, ... (j = 0
corresponds to the through state and represents the residual attenuation of the
device).
f) Compute estimates of the incremental attenuations Â1 (j) = |Ŝ21 (j)| − |Ŝ21 (0)| of
the error corrected measurements. |Ŝ21 (j)| and |Ŝ21 (0)| need to be in logarithmic
units. Repeat this process for S12 to calculate Â2 (j).
g) Take the larger of the two differences between measured and characterized incre-
mental attenuations: dj = max(|Â1 (j) − Aref (j)|, |Â2 (j) − Aref (j)|).
h) Define a reasonable envelope that encloses the calculated differences dj over the
whole attenuation range. Since the frequency dependence is generally weak, this
envelope might enclose any frequency dependence.
i) Transform dj to linear units using dj,lin ' 0.12 dj and use dj,lin as uncertainties
u(|L̂i |) and u(arg(L̂i )).
The procedure above is limited by the uncertainty associated with the attenuation
steps. These uncertainties should be taken into account, if significant. The stated non-
linearities can’t be smaller than the uncertainty associated with the attenuation steps.
Figures G.7, G.8 and G.9 illustrate the procedure and provide additional information.
1.5
Aˆ1 − Aref /dB
0.5
−0.5
−90 −80 −70 −60 −50 −40 −30 −20 −10
Aref /dB
Figure G.7: Characterization of VNA non-linearity according to procedure G.2. The dif-
ference between a measurement and the reference value of incremental attenuation is
shown for the entire attenuation range down to -90 dB at 1 GHz. This corresponds to
step g) of the procedure, although only showing data for S21 . The different colors indi-
cate four repetitions of the same measurement sequence. The large deviation from 0 and
enhanced scatter below -50 dB are due to the influence of noise floor. Therefore, for the
characterization of non-linearity only data above -60 dB should be considered, see figure
G.8
0.05
0.04
0.03
0.02
Aˆ1 − Aref /dB
0.01
−0.01
−0.02
−0.03
−0.04
−0.05
−55 −50 −45 −40 −35 −30 −25 −20 −15 −10 −5
Aref /dB
Figure G.8: The same as figure G.7. Only data above -60 dB should be used to char-
acterize VNA non-linearity. Uncertainty intervals are shown as well. They need to be
considered, when defining the envelope according to step h) of the procedure.
0.04
0.03
0.02
Aˆ1 − Aref /dB
0.01
−0.01
−0.02
−0.03
−0.04
−0.05
−55 −50 −45 −40 −35 −30 −25 −20 −15 −10 −5
Aref /dB
Figure G.9: The same as figure G.7. A single measurement sequence at three different
frequencies, 1 GHz (blue), 10 GHz (green) and 18 GHz (red) is shown. The data indicates
that the frequency dependence of non-linearity is weak. The orange line indicates a
possible envelope, according to step h).
b) Measure the states of the ECU repeatedly over a period of 24 hours with an inter-
val of 15 minutes. This necessitates computerized control of the ECU switching
states (refer to VNA manual, ECU manual or contact manufacturer).
c) Use each measurement set to determine estimates of the error coefficients of the
VNA as a function of time Êij (t), starting at t = 0.
d) Determine the drift of the error coefficients related to directivity and match and the
switch terms with respect to the first measurement at t = 0 with ∆Eij (t) = Êij (t) −
Êij (t = 0). Calculate the maximum deviation as max(| Re[∆Eij (t)]|, | Im[∆Eij (t)]|).
Define a reasonable envelope that encloses the maximum deviation over the 24
hour time span. This could e.g. be a linear function of time and/or frequency.
Take the values of this envelope as standard uncertainties associated with the
estimates of both the real and imaginary components of the drift terms. It is im-
portant that these drift functions are defined rather generously to enclose possible
variations in drift behavior.
e) Determine the drift of the error coefficients related to transmission with ∆Eij (t) =
Êij (t)/Êij (t = 0). Calculate the deviations in magnitude as |∆Eij (t)| − 1 and in
phase as | arg(∆Eij (t))| Define a reasonable envelope that encloses the devia-
tions over the 24 hour time span. This could e.g. be a linear function of time
and/or frequency. Take the values of these envelopes as standard uncertainties
associated with the estimates of magnitude and phase of the drift terms. It is im-
portant that these drift functions are defined rather generously to enclose possible
variations in drift behavior.
d) Determine the drift of S11 and S22 with respect to the first measurement at t = 0
with ∆Sii (t) = Ŝii (t) − Ŝii (t = 0). Calculate the maximum deviation as
max(| Re[∆S11 (t)]|, | Im[∆S11 (t)]|, | Re[∆S22 (t)]|, | Im[∆S22 (t)]|). Define a reason-
able envelope that encloses the maximum deviation over the 24 hour time span.
This could e.g. be a linear function of time and/or frequency. Take the values of
this envelope as standard uncertainties associated with the estimates of both real
and imaginary components of the drift terms related to the error coefficients direc-
tivity, match and switch terms. It is important that these drift functions are defined
rather generously to enclose possible variations in drift behavior.
e) Determine the drift of S21 and S12 with respect to the first measurement at t = 0
with ∆Sij (t) = Ŝij (t)/Ŝij (t = 0). Calculate the maximum deviations in magnitude
and phase as max(|∆S21 (t)|−1, |∆S12 (t)|−1) and max(| arg[∆S21 (t)]|, | arg[∆S12 (t)]|)
Define a reasonable envelope that encloses the maximum deviations over the 24
hour time span. This could e.g. be a linear function of time and/or frequency.
Take the values of these envelopes as standard uncertainties associated with the
estimates of magnitude and phase of the drift terms associated with the error
coefficients related to transmission. It is important that these drift functions are
defined rather generously to enclose possible variations in drift behavior.
Figures G.10, G.11 and G.12 illustrate the procedure without ECU. At very low fre-
quency the enhanced noise floor due to inefficient couplers is visible again, in particular,
in figures G.10 and G.11. This effect should already be taken into account in the char-
acterization of the VNA noise floor, described in G.1, and can be ignored in the drift
characterization.
Drift is affected by changes in the environment. After performing the VNA calibra-
tion, the operator connects the flush thru, starts the automated measurement sequence
and then leaves the setup alone. The effect of this handling at the start of the procedure
can be seen in figure G.12 when the phase of the transmission coefficient shows the
largest deviation at the beginning of the sample of observations. To a lesser extent it
can be also observed for the magnitude (figure G.11). Later in a quiet environment, the
changes become more gradual.
It is important to understand that the procedure above characterizes drift related to
the entire setup, consisting of VNA and test port cable. Using another test port cable
might provide different results.
4
Ŝii (t) − Ŝii (t = 0)
−2
−4
−6
0 10 20 30 40 50 60
Frequency /GHz
Figure G.10: Characterization of VNA drift with a flush thru measurement haccording
i to
the second procedure in G.3. The traces are repeated measurements of Re Ŝ11 (black),
h i h i h i
Im Ŝ11 (blue), Re Ŝ22 (green), Im Ŝ22 normalized to the first measurements, accord-
ing to step d) of the procedure. The measurements were done every 15 minutes for a
duration of 24 hours. The orange line indicates a possible envelope, according to step
d).
0.008
0.006
-Ŝij (t)/Ŝij (t = 0)- − 1
0.004
-
-
0.002
−0.002
−0.004
-
-
−0.006
−0.008
−0.01
0 10 20 30 40 50 60
Frequency /GHz
Figure G.11: Characterization of VNA drift with a flush thru measurement according
to
the second procedure in G.3. The traces are repeated measurements of Ŝ21 (green) and
Ŝ12 (blue) normalized to the first measurements, according to step e) of the procedure.
The measurements were done every 15 minutes for a duration of 24 hours. The orange
line indicates a possible envelope, according to step e).
1.5
1
arg Ŝij (t)/Ŝij (t = 0) /◦
2
0.5
−0.5
1
−1
−1.5
0 10 20 30 40 50 60
Frequency /GHz
Figure G.12: Characterization of VNA drift with a flush thru measurement according
to
the second procedure in G.3. The traces are repeated measurements of arg Ŝ21 (green)
and arg Ŝ12 (blue) normalized to the first measurements, according to step e) of the
procedure. The orange line indicates a possible envelope. The measurements were
done every 15 minutes for a duration of 24 hours.
b) Connect the test port cable to the VNA port. Wait for at least 30 minutes to achieve
thermal equilibrium. To avoid the waiting time see figures G.15 and G.16 for an
alternative setup.
d) Hold the cable in a straight line and take a first measurement (n = 1) of magnitude
and phase.
f) Take several measurements n = 2, 3, ... of magnitude and phase during the grad-
ual bend. Record the maximum deviation in magnitude and phase of the reflec-
tion coefficient with respect to measurement n = 1: ∆r = maxn (||Ŝ11 (n)/Ŝ11 (n =
1)| − 1|) and ∆φ = maxn (arg(Ŝ11 (n)/Ŝ11 (n = 1))). Define a generous envelope
over frequency, enclosing the maximum deviations.
g) Use the envelope values determined in the previous step as uncertainties that are
ˆ T |) =
associated with the estimate of the term CAT in figure F.5 leading to u(|CA
ˆ
∆r/2 and u(arg(CAT )) = ∆φ/2. The division by 2 for magnitude and phase is
applied, because the characterization with a short evaluates the stability of the
cable for a signal that propagates back and forth in the cable. The uncertainty
terms, however, are related to one-path effects.
h) For the term CAR in figure F.5 steps c) to g) have to be repeated with a matched
load connected to the VNA cable instead of a short. In this case, the measured
magnitude of S11 is small and hence the phase has minor significance. Thus,
an evaluation in real and imaginary components is preferable, leading to ∆x =
maxn (Re[Ŝ11 (n) − Ŝ11 (n = 1)]) and ∆y = maxn (Im[Ŝ11 (n) − Ŝ11 (n = 1)]). The
uncertainties assigned to the real and imaginary parts of an estimate of CAR are
ˆ R ) = u(Re[CA
u(CA ˆ R ]) = u(Im[CA ˆ R ]) = max(∆x, ∆y).
Figures G.13 and G.14 illustrate the characterization of the cable with respect to
transmission stability. The corresponding measurement setups are shown in figures
0.02
-Ŝ11 (n)/Ŝ11 (n = 1)- − 1
0.01
-
-
−0.01
-
-
−0.02
−0.03
0 10 20 30 40 50 60
Frequency /GHz
Figure G.13: Characterization of transmission magnitude stability of VNA test port cable
according to the procedure in G.4. The traces are repeated normalized measurements,
corresponding to step f) of the procedure, with the short connected to the end of the test
port cable. The orange line indicates a possible envelope. The measurement setup is
shown in figures G.15 (initial position of cable for measurement n = 1) and G.16 (final
position after gradual bend of cable).
4
arg Ŝ11 (n)/Ŝ11 (n = 1) /◦
2
−2
1
−4
−6
0 10 20 30 40 50 60
Frequency /GHz
Figure G.14: The same as figure G.13 but now for the phase.
Figure G.16: The same as figure G.15, but showing the end position for the characteriza-
tion of a VNA test port cable.
c) Repeat the previous step by re-connecting the short under different azimutal posi-
tions, at least 16 times to obtain a sample of estimates: (Ŝ11 (1) Ŝ11 (2) Ŝ11 (3) . . . )
d) Calculate the difference between maximum and minimum values of the sample in
the real component of Ŝ11 : ∆x = maxi (Re[Ŝ11 (i)]) − mini (Re[Ŝ11 (i)])
e) Calculate the difference between maximum and minimum values of the mea-
surement sample in the imaginary component of Ŝ11 : ∆y = maxi (Im[Ŝ11 (i)]) −
mini (Im[Ŝ11 (i)]).
g) Define a reasonable envelope that encloses ∆z/2 over the whole frequency range.
Typically, this will be done by dividing the whole frequency range into frequency
bands, each with a constant value that encloses the calculated values. This can
be taken as the standard uncertainties associated with the real and imaginary
components of CO ˆ R.
NOTE — From a statistical point of view it seems more appropriate to take the standard
deviation of the measurement sample as a measurement uncertainty, following the formalism in
C.5. instead of the interval enclosing all values. Taking the interval that encloses all values is a
more conservative approach, acknowledging the fact that this effect can strongly vary from one
connector pair to another within the same connector family.
Figure G.17 shows example data for the evaluation of connector repeatability for the
1.85 mm connector family. The data has been collected according to the procedure
above.
4
1 2
2
Ŝ11 − m Ŝ11
−2
−4
−6
0 10 20 30 40 50 60
Frequency /GHz
H.2 Uncertainties
The uncertainties derived in this context do not need to be interpreted as uncertain-
ties in the estimated magnitudes of the complex-valued quantities, as was the case
in previous versions of the guide. Instead, they can be seen as equal sized uncer-
tainties associated with the real and the imaginary components of the complex-valued
estimate of the S-parameter, assuming there is no correlation between them, e.g. for
S11 : u(Ŝ11 ) = u(Re[Ŝ11 ]) = u(Im[Ŝ11 ]). This corresponds to a circular uncertainty re-
gion in the complex plane, following the suggestions in [64]. This uncertainty can be
transformed to an uncertainty in magnitude and phase using standard uncertainty prop-
agation techniques.
Sc11 denotes the error corrected reflection coefficient. The residual error coefficients
correspond to those listed in table D.1. The drift terms Dij are omitted in equation (H.2).
They have to be considered as a part of the uncertainties associated with the estimates
of the residual error coefficients. The term L denotes non-linearity.
(H.3c)
h i2 h R i2 h R i2 h R i2
u Ŝ12
ˆ
= Ŝc22 Ŝc12 u δE
ˆ
+ Ŝc11 Ŝc12 u δE
ˆ
+ Ŝc12 u δE +
22 11 01
h R i2 h i2 h i2
u δE ˆ + Ŝc12 u L̂
+ Ŝc12 u Ĉ23
+
03
h i2 h i2
c c
Ŝ12 Ŝ22 u Ĉ22 + u R̂S12 .
(H.3d)
The terms Scij denote error corrected S-parameters. The residual error coefficients
δEFij of the forward model and δER ij of the reverse model correspond to those listed
in table D.2. Again, the drift terms are part of the uncertainties associated with the
estimates of the residual error coefficients, as in the one-port case. The terms Cij
denote cable movements on the port two side. The term L denotes non-linearity. The
terms RSij summarize all random effects.
H.4.2.1 Limitations of the residual measurement model
From practical experience it is known that uncertainties associated with transmission
measurement can be underestimated with equations (H.3b) and (H.3d), if the raw load
match of the VNA is significant. This is a shortcoming of the residual measurement
(H.4a)
h i2 h R i2 h R i2 h R i2
u Ŝ12 = Ŝc22 Ŝc12 u δE
ˆ ˆ
+ Ŝc11 Ŝc12 u δE
ˆ
+ Ŝc12 u δE +
22 11 01
h R i2 h i2 h i2
u δEˆ + Ŝc12 u L̂
+ Ŝc12 u Ĉ23
+
03
h i2 h i2
c c
Ŝ12 Ŝ22 u Ĉ22 + u R̂S12 +
2 h F i2 h R i2
c R
Ŝ Ê 1 − Ŝc Ŝc
u δE ˆ + u δE ˆ .
12 11 21 12 00 22
(H.4b)
Equations (H.4a) and (H.4b) each have one additional term, containing estimates of the
load matches, ÊF22 and ÊR
11 , of the respective ports.
0.03
0.025
0.02
|S11 |
0.015
0.01
0.005
0
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure H.1: The ripple pattern of a 30 cm beadless Type-N air-dielectric line terminated
with a matched load.
adjustment made for any slope caused by the variation with frequency of the reflection
coefficient of the terminating load, as illustrated in figure H.2. AD corresponds to an
estimate of twice the size of |δE00 |.
NOTE — The load must have a reflection coefficient ΓL that is larger than the magnitude
of the residual directivity, i.e. |ΓL | > |δE00 |. If this is not the case, the peak-to-peak ripple value
AD will be an estimate of 2 |ΓL |, rather than 2 |δE00 |. This would lead to an underestimation of
the residual directivity.
Figure H.2: A practical way to compensate for the slope when the peak-to-peak ripple
value AD is determined. AD is indicated by the red solid line.
with dpg the maximal pin gap in µm and f the frequency in GHz. The factor k is
dependent on the connector type, k = 8 · 10−5 for Type-N and k = 7 · 10−5 for 3.5 mm.
dpg is the absolute value of the sum of the recession of the center conductor at the test
port and the maximum recession of the center conductor of the air-dielectric line. The
uncertainty associated with |Γ̂CO | can be approximated by
u Γ̂CO ' 0.1kf · u dˆpg . (H.6)
Example: The recession of the center conductor is -20 µm at the test port, -20 µm at
the air-dielectric line and -10 µm at the matched load. This leads to a maximal pin gap of
dˆpg = 50 µm. For a measurement in Type-N at 18 GHz this leads to |Γ̂CO | ' 8.64 · 10−3
when using equation (H.5). A single measurement of the pin depth is assumed to have an
associated uncertainty of 4 µm. The addition of the three measurements leads to u(dˆpg ) =
12 µm, assuming that they are strongly correlated because the same connector gauge is used
for all three measurements. This leads to u(|Γ̂CO |) ' 1.73 · 10−3 when using equation (H.6).
NOTE — Equation (H.5) seems to favor a small maximal pin gap to keep the interfering
reflections as small as possible. However, caution is advised. Equation (H.5) takes into account
inductive coupling only. In case of a small pin gap capacitive coupling can become dominant
leading to stability problems, see 8.5.2.2. The size of this effect can’t be easily estimated an-
alytically. It is therefore necessary to repeat the ripple evaluation under different connector
orientations and record the variation in the ripple amplitude as described in H.5.1.4.
H.5.1.4 Repeatability
If more than one suitable air-dielectric line is available, the procedure should be re-
peated using the other available air-dielectric lines to check for the consistency of the
value obtained for the residual directivity. The technique should also be repeated sev-
eral times using the same air-dielectric line to estimate the variability in the value ob-
tained for the residual directivity. Typically only a small number of repetitions will be
The evaluations of the different terms in equation (H.9) are explained in H.5.1.1, H.5.1.2,
H.5.1.3 and H.5.1.4, respectively. The uncertainty associated √ with an estimate of the
drift u(|D̂00 |) is evaluated according to G.3. The divisor of 2 acknowledges the fact
that the phase is unknown, see [50, 69].
H.5.2 Source match
The uncertainties u(δE ˆ F ) and u(δE
ˆ 11 ), u(δE ˆ R ) are determined according to the evalua-
11 22
tion of uncertainties associated with the residual directivity, except that a high-reflecting
short is used instead of a matched load as termination of the air-dielectric line. The
residual source match is typically in the range 0.001 to 0.030 (i.e., from -60 dB to -
30 dB) [67].
The procedure below is limited to one-port measurements and the determination of
ˆ
u(δE11 ). For two-port measurements the same procedure is performed at each port to
determine u(δE ˆ F ) and u(δE ˆ R ).
11 22
0.99
0.98
0.97
0.96
|S11 |
0.95
0.94
0.93
0.92
0.91
0.9
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure H.3: Ripple pattern of a 30 cm beadless Type-N air-dielectric line terminated with
a short.
Example: The attenuation of AS due to losses can be estimated from the average re-
flection coefficient in figure H.3. At the maximum frequency the average reflection coefficient
is approximately 0.915. Because the observed residual source match scales with S211 the ob-
served AS needs to be multiplied with a factor cL = 1/0.9152 = 1.195 at this frequency.
(H.10)
The first term in equation (H.10) is determined according to H.5.2.1 with AS the peak-
to-peak value in the ripple evaluation and cL the amplification factor to compensate for
losses in the air-dielectric line. The second term acknowledges the influence of the
residual directivity with AD determined in H.5.1.1. For the evaluation of the remaining
terms the same procedures as described for the residual directivity (H.5.1.2, H.5.1.3
and H.5.1.4) can be applied. The uncertainty associated √ with the estimate of drift,
u(|D̂11 |), is evaluated according to G.3. The divisor of 2 takes into account that the
phase is unknown, see [50, 69].
with x denoting
Γ̂o
x=−− 1.
Γ̂s
The subscript o is referring to the calibration open. To apply equation (H.14) it is nec-
essary that short and open have well matched electrical lengths to keep the factor x
The procedures to assign uncertainties associated with the estimates of |CAT |, arg(CAT )
and CAR are given in G.4. For arg(CAT ) the uncertainty needs to be expressed in ra-
dians.
H.5.8 Non-linearity
The uncertainty associated with the estimate of L is determined according to procedure
G.2.
H.5.9 Repeatability
This uncertainty contribution includes all random effects related to the measurement of
the DUT. It is generally dependent on the settings of the VNA, influencing noise floor
and trace noise, and the connection between test port and DUT. It is possible to quote
an uncertainty associated with the estimate of repeatability based on experience with
DUTs of similar reflection and transmission coefficient and of similar type of construc-
tion. For this purpose pooled data can be analyzed according to the formalism and the
advice given in C.5. Based on that approach, a value can be assigned to uncertainties
associated with estimates of RSij . It is nevertheless always necessary to perform at
least four measurements at different connector orientations to verify the quoted uncer-
tainty, as pointed out in 8.8.
I.2 Equipment
I.2.1 VNA test ports
In order to make measurements in waveguide, the VNA needs to be configured so that
it has waveguide test ports of the correct aperture size and flange type for the required
measurements. Depending on the type of VNA, there are two ways to achieve this:
• Using coaxial cables and/or adapters to transform from coaxial tests ports found
either on the front panel of the VNA or on external coaxial Extender Heads (e.g.
as often used for 1 mm coaxial connector VNA test ports).
In both cases, it is very important that the waveguide test ports consist of good quality
waveguide that is in good condition.
I.3 Calibration
I.3.1 SSL calibration
A common one-port calibration technique is the flush-short / offset-short / well-matched
load technique. This is sometimes called the SSL calibration technique (short-short-
load).
The SSL calibration technique is analogous with the SOL (short-open-load) calibra-
tion technique that is used frequently in coaxial line. Sometimes, the SSL calibration
is implemented using two offset-shorts (i.e. by using a second offset-short in place of
the flush-short). This works well as long as the phases of the reflection coefficients
produced by the offset-shorts do not coincide with each other at any frequency across
the waveguide band.
SSL calibrations can be extended to two-port calibrations by applying a SSL calibra-
tion to each of the VNA’s two test ports followed by a Thru connection (made by joining
the two waveguide tests ports together). This produces a SSLT calibration (short-short-
load-thru).
10
7
|S11 |
2
8.5 9 9.5 10 10.5 11 11.5 12
Frequency /GHz
Figure I.1: Example of a residual directivity ‘ripple’ trace obtained from a VNA calibration
in X-band waveguide (operating from 8.2 GHz to 12.4 GHz).
1.005
0.995
|S11 |
0.99
0.985
0.98
8.5 9 9.5 10 10.5 11 11.5 12
Frequency /GHz
Figure I.2: Example of a residual source match ‘ripple’ trace obtained from a VNA cali-
bration in X-band waveguide (operating from 8.2 GHz to 12.4 GHz).
0.035
0.03
0.025
|S11 |
0.02
0.015
0.01
0.005
0
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.1: Four different ripple patterns in the Type-N line system. All four traces were
taken with a 30 cm beadless air-dielectric line terminated with a matched load. The dif-
ferences are due to different longitudinal positions of the center conductor of the air-
dielectric line. See also H.5.1.2
well. Results are expressed either in real and imaginary components, for low reflection
coefficients, or magnitude and phase, for the other cases. The magnitude of the trans-
mission coefficient is displayed in logarithmic units. The results are shown graphically
over the whole frequency range. For three selected frequencies, uncertainty budgets
are shown in tabular form. The uncertainty budget of the Ripple Method quotes a sin-
gle uncertainty u(Sij ) = u(Re[Sij ]) = u(Im[Sij ]), as explained in H.2. The uncertainties
associated with magnitude and phase are derived using linear uncertainty propagation
u(|Sij |) = u(Sij )
u(Sij )
u(arg (Sij )) = .
|Sij |
The transformation between linear and logarithmic units is also done using linear un-
certainty propagation
20 u(Xlin ) u(Xlin )
Xlog = 20 log10 Xlin −→ u(Xlog ) = ' 8.686
loge 10 Xlin Xlin
Xlog u(Xlin ) loge 10
Xlin = 10 20 −→ = u(Xlog ) ' 0.1151u(Xlog ) .
Xlin 20
The numerical values of the uncertainty contributions in these examples should not
be taken as generally representative. E.g. the uncertainty associated with the repeata-
bility of the DUT measurements is very small in these examples. Such uncertainties
can only be achieved with appropriate device settings, a well controlled measurement
process and stable environmental conditions. It is necessary to individually character-
ize the uncertainty contributions for the specific measurement setups. This can also be
0.015
0.01
0.005
0
Re[S11 ]
−0.005
−0.01
−0.015
−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.2: Real component of the reflection coefficient of a matched load with associ-
ated standard uncertainties. The solid lines indicate the error corrected measurement
data (center line) bounded by the interval of one standard uncertainty (k = 1) calculated
with the rigorous method. The step in uncertainty at 2 GHz is due to the change from
low band load to sliding load. The vertical bars indicate standard uncertainties evaluated
with the Ripple Method.
0.02
0.015
0.01
0.005
Im[S11 ]
−0.005
−0.01
−0.015
−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.3: The same as figure J.2 but for the imaginary component of the reflection
coefficient.
Table J.3: Uncertainty budget of S11 of a matched load at three different frequencies. The
standard uncertainty is evaluated with the Ripple Method. The terms in column 2 refer
to equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.
0.23
0.22
0.21
0.2
|S11 |
0.19
0.18
0.17
0.16
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.4: Magnitude of the reflection coefficient of a mismatch with associated stan-
dard uncertainties. The solid lines indicate the error corrected measurement data (center
line) bounded by the interval of one standard uncertainty (k = 1) calculated with the rigor-
ous method. The step in uncertainty at 2 GHz is due to the change from low band load to
sliding load. The vertical bars indicate standard uncertainties evaluated with the Ripple
Method.
1
arg (S11 ) /◦
−1
−2
−3
−4
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.5: The same as figure J.4 but for the phase of the reflection coefficient. For
better visibility the error corrected measurement has been normalized to 0 and only the
standard uncertainties are shown.
Table J.5: The same as table J.4 but for arg (S11 )
Table J.6: Uncertainty budget of S11 of a mismatch at three different frequencies. The
standard uncertainty is evaluated with the Ripple method. The terms in column 2 refer
to equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.
1.04
1.03
1.02
1.01
|S11 |
0.99
0.98
0.97
0.96
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.6: Magnitude of the reflection coefficient of a short with associated standard
uncertainties. The solid lines indicate the error corrected measurement data (center line)
bounded by the interval of one standard uncertainty (k = 1) calculated with the rigor-
ous method. The vertical bars indicate standard uncertainties evaluated with the Ripple
Method.
1.5
0.5
arg (S11 ) /◦
−0.5
−1
−1.5
−2
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.7: The same as figure J.6 but for the phase of the reflection coefficient. For
better visibility the error corrected measurement has been normalized to 0 and only the
standard uncertainties are shown.
Table J.8: The same as table J.7 but for arg (S11 )
Table J.9: Uncertainty budget of S11 of a short at three different frequencies. The stan-
dard uncertainty is evaluated with the Ripple method. The terms in column 2 refer to
equation (H.2) and the corresponding values in the subsequent columns have to be
added accordingly for the combined standard uncertainty.
0.025
0.02
0.015
0.01
Re[S22 ]
0.005
−0.005
−0.01
−0.015
−0.02
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.8: Real component of the reflection coefficient of an adapter with associated
standard uncertainties. The solid lines indicate the error corrected measurement data
(center line) bounded by the interval of one standard uncertainty (k = 1) calculated with
the rigorous method. The step in uncertainty at 2 GHz is due to the change from low band
load to sliding load. The vertical bars indicate standard uncertainties evaluated with the
Ripple Method.
0.03
0.02
0.01
Im[S22 ]
−0.01
−0.02
−0.03
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.9: The same as figure J.8 but for the imaginary component of the reflection
coefficient.
−0.05
−0.15
−0.2
−0.25
−0.3
−0.35
−0.4
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
0.8
0.6
0.4
0.2
arg (S12 ) /◦
−0.2
−0.4
−0.6
−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.11: The same as figure J.10 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.
Table J.14: The same as table J.13 but for arg (S12 )
0.06
0.04
0.02
0
Re[S22 ]
−0.02
−0.04
−0.06
−0.08
−0.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
0.06
0.04
0.02
0
Im[S22 ]
−0.02
−0.04
−0.06
−0.08
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.13: The same as figure J.12 but for the imaginary component of the reflection
coefficient.
−19.75
−19.8
−19.85
|S12 | /dB
−19.9
−19.95
−20
−20.05
−20.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
0.8
0.6
0.4
0.2
arg (S12 ) /◦
−0.2
−0.4
−0.6
−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.15: The same as figure J.14 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.
Table J.20: The same as table J.19 but for arg (S12 )
0.08
0.06
0.04
0.02
Re[S22 ]
−0.02
−0.04
−0.06
−0.08
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
0.06
0.04
0.02
0
Im[S22 ]
−0.02
−0.04
−0.06
−0.08
−0.1
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.17: The same as figure J.16 but for the imaginary component of the reflection
coefficient.
−49.4
−49.6
−50
−50.2
−50.4
−50.6
−50.8
−51
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
0.8
0.6
0.4
0.2
arg (S12 ) /◦
−0.2
−0.4
−0.6
−0.8
0 2 4 6 8 10 12 14 16 18
Frequency /GHz
Figure J.19: The same as figure J.18 but for the phase of the transmission coefficient.
For better visibility the error corrected measurement has been normalized to 0 and only
the standard uncertainties are shown.
Table J.26: The same as table J.25 but for arg (S12 )